Showing a dict can be slow

[ncnc]

The showkv function in dict.jl (latest master) contains the following code:

    for (i, v) in enumerate(iter)
        print(io, "\n  ")
        limit && i >= rows && (print(io, "⋮"); break)

        str = sprint(show, v)
        limit && (str = _truncate_at_width_or_chars(str, cols, "\r\n"))
        print(io, str)
    end

If v is "large", the above can be arbitrarily slow due to the sprint(show, v) part.

Below an example showing the problem from the REPL:

N = 1000000
d = Dict{ASCIIString,Array{Float64}}()
for i = 1:50
    d["Label $(i)"] = rand(N,1)
end
d   # takes a while to print

Increase N as necessary.

[mbauman]

Yup. This was my doing... and was why I then opened issue #5709 to try to figure out a consistent API to ask for a limited sprint. I'd be open to something better here, but I certainly prefer this to printing the whole v string.

[vtjnash]

we could create an IOBuffer with a max length, and then catch the write error if it happens. not sure if that is the best solution, however.

[ncnc]

@mbauman I agree that #5709 should cover this once it gets done. We definitely don't want to print the whole thing.

@vtjnash Interesting approach, if this issue starts bugging me enough I might give it a shot at some stage.

[hayd]

@vtjnash except for the failing early there's already a maxsize for IOBuffer (at least I don't think this fails early?). In fact there's already a sprint with a size option, though it's unclear to me if the intention was for this to be a sizehint or a maxsize:

s = IOBuffer(Array(UInt8,size), true, true, size)  # would make it a maxsize

[JeffBezanson]

This works pretty well as a stop-gap:

diff --git a/base/dict.jl b/base/dict.jl
index a496b5d..1fed3f5 100644
--- a/base/dict.jl
+++ b/base/dict.jl
@@ -97,9 +97,11 @@ function showdict{K,V}(io::IO, t::Associative{K,V}; limit::Bo
         print(io, key)
         print(io, " => ")

-        val = sprint(show, v)
         if limit
+            val = with_output_limit(()->sprint(show, v))
             val = _truncate_at_width_or_chars(val, cols - keylen, "\r\n")
+        else
+            val = sprint(show, v)
         end
         print(io, val)
     end